Bars are for categorical comparison: the eye compares bar endpoints, so it's easy to see which category is largest. Because we judge by length, bar charts must start at zero — otherwise the lengths lie.
🔎 See it for yourself — the zero-baseline trap
Quarterly sales: RM102k → 105k → 108k → 110k. Here's a chart with the axis starting at RM100k…
102Q1
105Q2
108Q3
110Q4
😱 “Sales exploded — Q4 is 5× Q1!” … but is it really?
From a zero baseline, the honest story appears.
102Q1
105Q2
108Q3
110Q4
Same numbers — a gentle 8% rise, not a fivefold jump. The truncated axis was the lie.
● Key idea
Bar charts start at zero — always. (This rule does not apply to line charts, which show shape, not length.) Also keep bars wider than the gaps between them.
4.2 The bar family
Vertical / clustered — compare categories, or several series side by side (e.g. sales by branch each quarter).
Stacked — show a total and its parts; but only the bottom segment shares a baseline, so parts above are hard to compare.
Horizontal — best when category names are long (they read left-to-right).
Stacked horizontal — ideal for Likert survey data (strongly disagree → strongly agree).
● Example · Likert survey
Course-feedback responses (“The labs were useful”: strongly disagree → strongly agree) sit naturally on a stacked horizontal bar — one bar per question, coloured by response, so you see the balance of opinion at a glance.
The bar family — same idea, different jobs
Clustered
Compare two series side by side.
Stacked
Total and its parts.
Stacked horizontal (Likert)
Balance of opinion per question.
Clustered compares series; stacked shows composition; stacked-horizontal is perfect for survey (Likert) data — disagree, neutral, agree.
● Session 1 recap
Bars compare categories, and the eye judges by length — so a bar chart must start at zero, or small differences look dramatic. Pick the right member of the bar family: clustered to compare series, stacked for totals, horizontal for long labels, stacked-horizontal for Likert data.
4.3 Histogram — the shape of one variable
A histogram groups numbers into bins and shows how many values fall in each. Unlike a bar chart, the bars touch, because the scale is continuous. Use it to see the distribution — normal, skewed, or with outliers.
● Example · Exam-mark distribution
Bin final marks into 0–20, 20–40, … 80–100 and count students per bin. One look tells you whether the class clusters around the middle (normal), leans low (left-skewed), or has a lonely high-scoring tail.
Histogram — distribution of final marks
30–20
920–40
2240–60
2860–80
1180–100
Bars touch (the scale is continuous). The class clusters in the 60–80 bin — a roughly normal, slightly left-leaning distribution. Number = students per bin.
4.4 Boxplot — a five-number summary you can see
A boxplot draws the minimum, Q1, median, Q3 and maximum, with outliers as separate dots. It's the fastest way to compare the spread of several groups at once.
● Connect it back
You already compute the five-number summary by hand. A boxplot is simply that summary drawn — and side-by-side boxplots let you compare, say, CGPA spread across three diploma programmes in one figure. Judge shape from the box, not the whiskers (outliers stretch whiskers).
Boxplots — CGPA spread across three programmes
Each box is a five-number summary: whisker caps = min/max, box = Q1–Q3, line = median. Programme B has the highest median and one low outlier; C is the most spread out.
4.5 Geospatial — data on a map
When location is the point, put the data on a map. A choropleth shades each region by value (darker = higher); bubbles size each place by value.
● Example · Dengue by state
Shade each Malaysian state by dengue cases — the hotspots (often Selangor and the Klang Valley) light up immediately, far faster than a 16-row table of state counts.
Choropleth — dengue cases by state, Malaysia
fewermore cases
Data on a real map: each state is shaded darker for more dengue cases, so Selangor and KL (the Klang Valley) read as the clear hotspot instantly — far faster than a 16-row table of state counts. (State boundaries: simplemaps.com.)
4.6 Cautious elements — how good data lies
Pie traps — too many slices, or 3D/shadow effects that distort sizes. Prefer bars for comparison.
Never use 3D for a single dimension — it warps values and adds nothing.
Avoid a secondary y-axis — two scales on one chart confuse; split into two charts instead.
Cut clutter — heavy gridlines, background shading, rainbow palettes and bold everywhere all raise effort and lower trust.
● Watch out
The two biggest lies in everyday charts are the truncated bar axis (Session 1) and the 3D pie. Both make small differences look dramatic. As a data-literate student, you should catch them instantly.
● Session 2 recap
A histogram shows the distribution of one variable (bins that touch); a boxplot draws the five-number summary to compare groups; geospatial charts map data by place. Avoid the traps: 3D, secondary y-axes and clutter — and never a truncated bar axis.
Key terms & definitions
Zero baseline
A bar chart's y-axis must start at zero, so bar lengths compare honestly.
Clustered / stacked bar
Clustered compares series side by side; stacked shows a total and its parts.
Likert (stacked horizontal)
A survey scale (strongly disagree → agree) shown as one stacked horizontal bar per question.
Histogram & bins
Groups numeric values into intervals (bins) and shows how many fall in each; bars touch.
Boxplot / five-number summary
Draws min, Q1, median, Q3, max (+ outliers) to show spread and compare groups.
Choropleth
A map shading each region by value — darker = higher.
Chartjunk
Decorative clutter that lowers clarity and trust — 3D, heavy gridlines, rainbow palettes.
✅ Quick self-check
1. A bar chart’s y-axis starts at 100 instead of 0. This most likely…
2. Which chart shows the distribution of one numeric variable?